Our Master’s Thesis series continues! Meet Anastasis Litsas, a gifted Software Engineer here at Ballista Technology Group. Check out how the latest advancements in Machine/Deep Learning paved the way for his thesis: Characterization and Control of Surface Structures in Laser-based Processing using Machine/Deep Learning.
What is the first idea that comes to mind when you hear about Deep Learning/Machine Learning? Some of you may think about classification problems of cats and dogs, and some may think about regression tasks or clustering. During my bachelor’s degree studies, I never thought there was something else I could lay my hands on and do for research within my industry. However, new industry innovations have created opportunities within the field of computer science that previously did not exist. For example, DALL-E is a generative model that makes AI-generated images. Or WaveNet, which generates speech as output with millions of trained parameters. These kinds of products seemed to need an extensive research team behind them and, most importantly, an idea. When I started my master’s degree, I was offered an opportunity to work on a project where I would research and train a generative Deep Learning model to generate, based on a laser parameters configuration, an image illustrating the material produced after the laser process. The need for such a product comes when you understand its potential. As labs need to simulate hundreds of costly experiments of laser processing materials to create unique or desired properties, my work would eventually create an inexpensive solution as it zeroed down both the time and cost of this process. Many paths have been taken to solve this problem. Until now, I’ve worked with Convolutional Neural Networks, AutoEncoders, and GANS, trying to utilize state-of-the-art and novel models such as Cumulant Gans. Despite the fact I have a long road ahead, having the chance to research and contribute to cutting-edge technology is invaluable!